Make Claude read your chat logs so you don't have to

What it does

mcp-server-chatsum is an MCP server that lets LLM clients like Claude Desktop rummage through your chat history. It exposes one tool, query_chat_messages, which searches a local chat database and returns summaries based on your prompt. Before it can do anything useful, you must run a bundled chatbot component to populate that database—this server only reads what has already been saved.

The interesting bit

The project is mostly plumbing: a thin translation layer between a chat log and the Model Context Protocol. The value is in turning unstructured banter into queryable context an LLM can access directly.

Key highlights

• Bundled chatbot directory generates and stores messages locally in a .db file
• Single MCP tool (query_chat_messages) handles both search and summarization
• Reads from a local database pointed to by the CHAT_DB_PATH environment variable
• Connects to Claude Desktop over stdio using standard MCP configuration
• TypeScript implementation with a conventional pnpm build setup

Caveats

• The README never explains how messages are summarized—whether by local inference, remote API, or prompt trickery is left unsaid
• Setup requires configuring the bundled chatbot first, but the documentation for that step is essentially “see the other README”

Verdict

Handy if you already live inside Claude Desktop and need to excavate old conversations for buried decisions. If you want a self-contained chat summarizer, look elsewhere—this is strictly bridge territory.